DuoNeural/Gemma4-31B-IT-Abliterated
Abliterated vs google/gemma-4-31B-it. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 58 of 60 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, Norms 0/523, other 0/62, MLP 58/261, Attention 57/338
2 derivatives checked
- mradermacher/Gemma4-31B-IT-Abliterated-i1-GGUF: Quantization (fewer layers)
- mradermacher/Gemma4-31B-IT-Abliterated-GGUF: Quantization (fewer layers)